2019_programme: APPLICATION OF MARKOV-CHAIN MONTE CARLO PARAMETER-SET IDENTIFICATION TO DESIGN AND INVERSION

• Session: 24. Inversion methods in underwater acoustics
  Organiser(s): N/A
• Lecture: APPLICATION OF MARKOV-CHAIN MONTE CARLO PARAMETER-SET IDENTIFICATION TO DESIGN AND INVERSION
  Paper ID: 744
  Author(s): Ivansson Sven
  Presenter: Ivansson Sven
  Presentation type: oral
  Abstract: In this paper, Markov-chain Monte Carlo (MCMC) sampling is used to identify sets in parameter space defined by the smallness of some "misfit" function. Two applications are given. The first concerns design of Alberich anechoic coatings with identification of coatings achieving a given maximum reflectance within a given frequency band. The layer-multiple-scattering (LMS) computational method is extended to cover three or more cavity types in the rubber coating. Three types turn out to improve the broad-band echo reduction, but no further improvement is obtained with four. Correlations among favourable design parameters are identified. The preferable cavity sizes typically increase with the rubber shear-wave velocity, but the smallest cavities are smaller than expected from a consideration of the resonance frequency of an isolated cavity. Effects of multiple scattering may explain this. The second application is about inversion uncertainty, with an example concerning determination of seismo-acoustic material parameters from backscattering measurements. With the a priori information that the difference between data and replica be within a certain error-norm bound, marginal estimates of the model parameters are computed. An interpretation is given in terms of Bayesian inversion with a certain simple data-error distribution, that does not involve additional data-error covariance parameters to be estimated.
    Download the full paper
• Corresponding author: Dr Ivansson Sven
  Affiliation: N/A
  Country: Sweden
  e-mail: